BGLab/BioTrove-Train

BioTrove是一个大型图像数据集，旨在通过人工智能技术促进生物多样性研究。该数据集包含大量经过处理的图像和元数据，支持图像分类和零样本分类任务。数据集分为多个子集，如BioTrove-Train（包含40M图像样本和33K物种）、BioTrove-Balanced（平衡物种分布）、BioTrove-Unseen（评估模型对未见物种的泛化能力）和BioTrove-LifeStages（评估模型对昆虫不同发育阶段的识别能力）。此外，数据集还提供了详细的元数据信息和软件工具，方便用户下载、访问和操作数据。

BioTrove is a large curated image dataset designed to enable AI for biodiversity research. The dataset includes detailed metadata and image URLs, covering multiple taxonomic groups such as Aves, Arachnida, Insecta, Plantae, Fungi, Mollusca, and Reptilia. Additionally, several sub-datasets are provided, such as BioTrove-Balanced, BioTrove-Unseen, and BioTrove-LifeStages, for various research purposes, including balanced species distribution, evaluating model generalization capability on unseen species, and recognizing species across different developmental stages. The dataset also offers related software tools and models, such as the BioTrove-CLIP model, for image classification and zero-shot classification tasks.